Small question: How do databases fit into using CRDTs? Are CRDTs normalized, storable inside the database somehow, allowing queries on them? Or do database completely fall on the wayside? This confuses me.
A CRDT's opaque blobs have to be treated as the source of truth for your data, because that's the only way they can resolve edits from multiple sources.
You could store an extracted version of the same data in your DB, alongside the blob, but it would have to be a read-only copy, with all edits routed through your CRDT library.
> Are CRDTs normalized
There are many kinds of CRDTs, supporting everything from single strings all the way up through arbitrary data structures.
You could model data inside your CRDT as either normalized or or denormalized. For some use cases, keeping a separate CRDT per record could work fine. But if you had e.g., a re-orderable todo list you'd want each todo to be part of a larger CRDT so that if two devices reorder records they both arrive at a single convergent ordering.
> do database completely fall on the wayside?
Not at all - DBs solve a lot of problems that CRDTs don't really touch. Robust data persistence, ACID, enforcing data rules, issuing arbitrary queries, etc.
CRDTs get multiple nodes to agree on a single view of the world, and many applications suitable for CRDTs would want to use them in conjunction with a database.
A database asserts a single, canonical, authoritative definition of state in a given domain. CRDTs assert multiple, concurrent, equivalent definitions of state in a given domain. In short, the semantics you get from a normal DB are stronger than the semantics you get from a normal CRDT, and the mapping between them is not one-to-one.
